Tools » teuthology

We do not currently have a way to detect whether memory use for daemons is remaining within acceptable bounds other than seeing runs on vps fail due to memory pressure. In such cases, it's almost certainly the case that the vps in under provisioned, making it a poor test.

Instead, update the ceph task to track daemon (osd, mon, mds) memory usage over the course of a run (the lifetime of the ceph task, that is) and add an option to fail the test if a particular daemon exceeds a configured amount of memory (will need different threshholds for osd, mon, and mds).

Second, we need such a threshhold. As a first pass, we should look at the stats from a week of runs and set the threshhold at 1.5x the mean max memory used by the daemon type. We'll probably need to adjust particular tests from there.